1 /*-------------------------------------------------------------------------
5 * The WAL sender process (walsender) is new as of Postgres 9.0. It takes
6 * care of sending XLOG from the primary server to a single recipient.
7 * (Note that there can be more than one walsender process concurrently.)
8 * It is started by the postmaster when the walreceiver of a standby server
9 * connects to the primary server and requests XLOG streaming replication.
10 * It attempts to keep reading XLOG records from the disk and sending them
11 * to the standby server, as long as the connection is alive (i.e., like
12 * any backend, there is a one-to-one relationship between a connection
13 * and a walsender process).
15 * Normal termination is by SIGTERM, which instructs the walsender to
16 * close the connection and exit(0) at next convenient moment. Emergency
17 * termination is by SIGQUIT; like any backend, the walsender will simply
18 * abort and exit on SIGQUIT. A close of the connection and a FATAL error
19 * are treated as not a crash but approximately normal termination;
20 * the walsender will exit quickly without sending any more XLOG records.
22 * If the server is shut down, postmaster sends us SIGUSR2 after all
23 * regular backends have exited and the shutdown checkpoint has been written.
24 * This instruct walsender to send any outstanding WAL, including the
25 * shutdown checkpoint record, and then exit.
28 * Portions Copyright (c) 2010-2011, PostgreSQL Global Development Group
31 * src/backend/replication/walsender.c
33 *-------------------------------------------------------------------------
41 #include "access/xlog_internal.h"
42 #include "catalog/pg_type.h"
43 #include "libpq/libpq.h"
44 #include "libpq/pqformat.h"
45 #include "libpq/pqsignal.h"
46 #include "miscadmin.h"
47 #include "replication/basebackup.h"
48 #include "replication/replnodes.h"
49 #include "replication/walprotocol.h"
50 #include "replication/walsender.h"
51 #include "storage/fd.h"
52 #include "storage/ipc.h"
53 #include "storage/pmsignal.h"
54 #include "tcop/tcopprot.h"
55 #include "utils/builtins.h"
56 #include "utils/guc.h"
57 #include "utils/memutils.h"
58 #include "utils/ps_status.h"
59 #include "utils/resowner.h"
62 /* Array of WalSnds in shared memory */
63 WalSndCtlData *WalSndCtl = NULL;
65 /* My slot in the shared memory array */
66 static WalSnd *MyWalSnd = NULL;
69 bool am_walsender = false; /* Am I a walsender process ? */
71 /* User-settable parameters for walsender */
72 int max_wal_senders = 0; /* the maximum number of concurrent walsenders */
73 int WalSndDelay = 200; /* max sleep time between some actions */
76 * These variables are used similarly to openLogFile/Id/Seg/Off,
77 * but for walsender to read the XLOG.
79 static int sendFile = -1;
80 static uint32 sendId = 0;
81 static uint32 sendSeg = 0;
82 static uint32 sendOff = 0;
85 * How far have we sent WAL already? This is also advertised in
86 * MyWalSnd->sentPtr. (Actually, this is the next WAL location to send.)
88 static XLogRecPtr sentPtr = {0, 0};
90 /* Flags set by signal handlers for later service in main loop */
91 static volatile sig_atomic_t got_SIGHUP = false;
92 volatile sig_atomic_t walsender_shutdown_requested = false;
93 volatile sig_atomic_t walsender_ready_to_stop = false;
96 static void WalSndSigHupHandler(SIGNAL_ARGS);
97 static void WalSndShutdownHandler(SIGNAL_ARGS);
98 static void WalSndQuickDieHandler(SIGNAL_ARGS);
99 static void WalSndXLogSendHandler(SIGNAL_ARGS);
100 static void WalSndLastCycleHandler(SIGNAL_ARGS);
102 /* Prototypes for private functions */
103 static bool HandleReplicationCommand(const char *cmd_string);
104 static int WalSndLoop(void);
105 static void InitWalSnd(void);
106 static void WalSndHandshake(void);
107 static void WalSndKill(int code, Datum arg);
108 static void XLogRead(char *buf, XLogRecPtr recptr, Size nbytes);
109 static bool XLogSend(char *msgbuf, bool *caughtup);
110 static void CheckClosedConnection(void);
111 static void IdentifySystem(void);
112 static void StartReplication(StartReplicationCmd * cmd);
115 /* Main entry point for walsender process */
119 MemoryContext walsnd_context;
121 if (RecoveryInProgress())
123 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
124 errmsg("recovery is still in progress, can't accept WAL streaming connections")));
126 /* Create a per-walsender data structure in shared memory */
130 * Create a memory context that we will do all our work in. We do this so
131 * that we can reset the context during error recovery and thereby avoid
132 * possible memory leaks. Formerly this code just ran in
133 * TopMemoryContext, but resetting that would be a really bad idea.
135 * XXX: we don't actually attempt error recovery in walsender, we just
136 * close the connection and exit.
138 walsnd_context = AllocSetContextCreate(TopMemoryContext,
140 ALLOCSET_DEFAULT_MINSIZE,
141 ALLOCSET_DEFAULT_INITSIZE,
142 ALLOCSET_DEFAULT_MAXSIZE);
143 MemoryContextSwitchTo(walsnd_context);
145 /* Set up resource owner */
146 CurrentResourceOwner = ResourceOwnerCreate(NULL, "walsender top-level resource owner");
148 /* Unblock signals (they were blocked when the postmaster forked us) */
149 PG_SETMASK(&UnBlockSig);
151 /* Tell the standby that walsender is ready for receiving commands */
152 ReadyForQuery(DestRemote);
154 /* Handle handshake messages before streaming */
157 /* Initialize shared memory status */
159 /* use volatile pointer to prevent code rearrangement */
160 volatile WalSnd *walsnd = MyWalSnd;
162 SpinLockAcquire(&walsnd->mutex);
163 walsnd->sentPtr = sentPtr;
164 SpinLockRelease(&walsnd->mutex);
167 /* Main loop of walsender */
172 * Execute commands from walreceiver, until we enter streaming mode.
175 WalSndHandshake(void)
177 StringInfoData input_message;
178 bool replication_started = false;
180 initStringInfo(&input_message);
182 while (!replication_started)
186 WalSndSetState(WALSNDSTATE_STARTUP);
187 set_ps_display("idle", false);
189 /* Wait for a command to arrive */
190 firstchar = pq_getbyte();
193 * Emergency bailout if postmaster has died. This is to avoid the
194 * necessity for manual cleanup of all postmaster children.
196 if (!PostmasterIsAlive(true))
200 * Check for any other interesting events that happened while we
206 ProcessConfigFile(PGC_SIGHUP);
209 if (firstchar != EOF)
212 * Read the message contents. This is expected to be done without
213 * blocking because we've been able to get message type code.
215 if (pq_getmessage(&input_message, 0))
216 firstchar = EOF; /* suitable message already logged */
219 /* Handle the very limited subset of commands expected in this phase */
222 case 'Q': /* Query message */
224 const char *query_string;
226 query_string = pq_getmsgstring(&input_message);
227 pq_getmsgend(&input_message);
229 if (HandleReplicationCommand(query_string))
230 replication_started = true;
235 /* standby is closing the connection */
239 /* standby disconnected unexpectedly */
241 (errcode(ERRCODE_PROTOCOL_VIOLATION),
242 errmsg("unexpected EOF on standby connection")));
247 (errcode(ERRCODE_PROTOCOL_VIOLATION),
248 errmsg("invalid standby handshake message type %d", firstchar)));
264 * Reply with a result set with one row, two columns. First col is system
265 * ID, and second is timeline ID
268 snprintf(sysid, sizeof(sysid), UINT64_FORMAT,
269 GetSystemIdentifier());
270 snprintf(tli, sizeof(tli), "%u", ThisTimeLineID);
272 /* Send a RowDescription message */
273 pq_beginmessage(&buf, 'T');
274 pq_sendint(&buf, 2, 2); /* 2 fields */
277 pq_sendstring(&buf, "systemid"); /* col name */
278 pq_sendint(&buf, 0, 4); /* table oid */
279 pq_sendint(&buf, 0, 2); /* attnum */
280 pq_sendint(&buf, TEXTOID, 4); /* type oid */
281 pq_sendint(&buf, -1, 2); /* typlen */
282 pq_sendint(&buf, 0, 4); /* typmod */
283 pq_sendint(&buf, 0, 2); /* format code */
286 pq_sendstring(&buf, "timeline"); /* col name */
287 pq_sendint(&buf, 0, 4); /* table oid */
288 pq_sendint(&buf, 0, 2); /* attnum */
289 pq_sendint(&buf, INT4OID, 4); /* type oid */
290 pq_sendint(&buf, 4, 2); /* typlen */
291 pq_sendint(&buf, 0, 4); /* typmod */
292 pq_sendint(&buf, 0, 2); /* format code */
295 /* Send a DataRow message */
296 pq_beginmessage(&buf, 'D');
297 pq_sendint(&buf, 2, 2); /* # of columns */
298 pq_sendint(&buf, strlen(sysid), 4); /* col1 len */
299 pq_sendbytes(&buf, (char *) &sysid, strlen(sysid));
300 pq_sendint(&buf, strlen(tli), 4); /* col2 len */
301 pq_sendbytes(&buf, (char *) tli, strlen(tli));
304 /* Send CommandComplete and ReadyForQuery messages */
305 EndCommand("SELECT", DestRemote);
306 ReadyForQuery(DestRemote);
307 /* ReadyForQuery did pq_flush for us */
314 StartReplication(StartReplicationCmd * cmd)
319 * Let postmaster know that we're streaming. Once we've declared us as
320 * a WAL sender process, postmaster will let us outlive the bgwriter and
321 * kill us last in the shutdown sequence, so we get a chance to stream
322 * all remaining WAL at shutdown, including the shutdown checkpoint.
323 * Note that there's no going back, and we mustn't write any WAL records
326 MarkPostmasterChildWalSender();
329 * Check that we're logging enough information in the WAL for
332 * NOTE: This only checks the current value of wal_level. Even if the
333 * current setting is not 'minimal', there can be old WAL in the pg_xlog
334 * directory that was created with 'minimal'. So this is not bulletproof,
335 * the purpose is just to give a user-friendly error message that hints
336 * how to configure the system correctly.
338 if (wal_level == WAL_LEVEL_MINIMAL)
340 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
341 errmsg("standby connections not allowed because wal_level=minimal")));
343 /* Send a CopyBothResponse message, and start streaming */
344 pq_beginmessage(&buf, 'W');
345 pq_sendbyte(&buf, 0);
346 pq_sendint(&buf, 0, 2);
351 * Initialize position to the received one, then the xlog records begin to
352 * be shipped from that position
354 sentPtr = cmd->startpoint;
358 * Execute an incoming replication command.
361 HandleReplicationCommand(const char *cmd_string)
363 bool replication_started = false;
366 MemoryContext cmd_context;
367 MemoryContext old_context;
369 elog(DEBUG1, "received replication command: %s", cmd_string);
371 cmd_context = AllocSetContextCreate(CurrentMemoryContext,
372 "Replication command context",
373 ALLOCSET_DEFAULT_MINSIZE,
374 ALLOCSET_DEFAULT_INITSIZE,
375 ALLOCSET_DEFAULT_MAXSIZE);
376 old_context = MemoryContextSwitchTo(cmd_context);
378 replication_scanner_init(cmd_string);
379 parse_rc = replication_yyparse();
382 (errcode(ERRCODE_SYNTAX_ERROR),
383 (errmsg_internal("replication command parser returned %d",
386 cmd_node = replication_parse_result;
388 switch (cmd_node->type)
390 case T_IdentifySystemCmd:
394 case T_StartReplicationCmd:
395 StartReplication((StartReplicationCmd *) cmd_node);
397 /* break out of the loop */
398 replication_started = true;
401 case T_BaseBackupCmd:
402 SendBaseBackup((BaseBackupCmd *) cmd_node);
404 /* Send CommandComplete and ReadyForQuery messages */
405 EndCommand("SELECT", DestRemote);
406 ReadyForQuery(DestRemote);
407 /* ReadyForQuery did pq_flush for us */
412 (errcode(ERRCODE_PROTOCOL_VIOLATION),
413 errmsg("invalid standby query string: %s", cmd_string)));
417 MemoryContextSwitchTo(old_context);
418 MemoryContextDelete(cmd_context);
420 return replication_started;
424 * Check if the remote end has closed the connection.
427 CheckClosedConnection(void)
429 unsigned char firstchar;
432 r = pq_getbyte_if_available(&firstchar);
435 /* unexpected error or EOF */
437 (errcode(ERRCODE_PROTOCOL_VIOLATION),
438 errmsg("unexpected EOF on standby connection")));
443 /* no data available without blocking */
447 /* Handle the very limited subset of commands expected in this phase */
451 * 'X' means that the standby is closing down the socket.
458 (errcode(ERRCODE_PROTOCOL_VIOLATION),
459 errmsg("invalid standby closing message type %d",
464 /* Main loop of walsender process */
468 char *output_message;
469 bool caughtup = false;
472 * Allocate buffer that will be used for each output message. We do this
473 * just once to reduce palloc overhead. The buffer must be made large
474 * enough for maximum-sized messages.
476 output_message = palloc(1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE);
478 /* Loop forever, unless we get an error */
482 * Emergency bailout if postmaster has died. This is to avoid the
483 * necessity for manual cleanup of all postmaster children.
485 if (!PostmasterIsAlive(true))
488 /* Process any requests or signals received recently */
492 ProcessConfigFile(PGC_SIGHUP);
496 * When SIGUSR2 arrives, we send all outstanding logs up to the
497 * shutdown checkpoint record (i.e., the latest record) and exit.
499 if (walsender_ready_to_stop)
501 if (!XLogSend(output_message, &caughtup))
504 walsender_shutdown_requested = true;
507 /* Normal exit from the walsender is here */
508 if (walsender_shutdown_requested)
510 /* Inform the standby that XLOG streaming was done */
511 pq_puttextmessage('C', "COPY 0");
518 * If we had sent all accumulated WAL in last round, nap for the
519 * configured time before retrying.
524 * Even if we wrote all the WAL that was available when we started
525 * sending, more might have arrived while we were sending this
526 * batch. We had the latch set while sending, so we have not
527 * received any signals from that time. Let's arm the latch
528 * again, and after that check that we're still up-to-date.
530 ResetLatch(&MyWalSnd->latch);
532 if (!XLogSend(output_message, &caughtup))
534 if (caughtup && !got_SIGHUP && !walsender_ready_to_stop && !walsender_shutdown_requested)
537 * XXX: We don't really need the periodic wakeups anymore,
538 * WaitLatchOrSocket should reliably wake up as soon as
539 * something interesting happens.
543 WaitLatchOrSocket(&MyWalSnd->latch, MyProcPort->sock,
544 WalSndDelay * 1000L);
547 /* Check if the connection was closed */
548 CheckClosedConnection();
552 /* Attempt to send the log once every loop */
553 if (!XLogSend(output_message, &caughtup))
557 /* Update our state to indicate if we're behind or not */
558 WalSndSetState(caughtup ? WALSNDSTATE_STREAMING : WALSNDSTATE_CATCHUP);
562 * Get here on send failure. Clean up and exit.
564 * Reset whereToSendOutput to prevent ereport from attempting to send any
565 * more messages to the standby.
567 if (whereToSendOutput == DestRemote)
568 whereToSendOutput = DestNone;
571 return 1; /* keep the compiler quiet */
574 /* Initialize a per-walsender data structure for this walsender process */
581 * WalSndCtl should be set up already (we inherit this by fork() or
582 * EXEC_BACKEND mechanism from the postmaster).
584 Assert(WalSndCtl != NULL);
585 Assert(MyWalSnd == NULL);
588 * Find a free walsender slot and reserve it. If this fails, we must be
589 * out of WalSnd structures.
591 for (i = 0; i < max_wal_senders; i++)
593 /* use volatile pointer to prevent code rearrangement */
594 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
596 SpinLockAcquire(&walsnd->mutex);
598 if (walsnd->pid != 0)
600 SpinLockRelease(&walsnd->mutex);
606 * Found a free slot. Reserve it for us.
608 walsnd->pid = MyProcPid;
609 MemSet(&walsnd->sentPtr, 0, sizeof(XLogRecPtr));
610 walsnd->state = WALSNDSTATE_STARTUP;
611 SpinLockRelease(&walsnd->mutex);
612 /* don't need the lock anymore */
613 OwnLatch((Latch *) &walsnd->latch);
614 MyWalSnd = (WalSnd *) walsnd;
619 if (MyWalSnd == NULL)
621 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
622 errmsg("number of requested standby connections "
623 "exceeds max_wal_senders (currently %d)",
626 /* Arrange to clean up at walsender exit */
627 on_shmem_exit(WalSndKill, 0);
630 /* Destroy the per-walsender data structure for this walsender process */
632 WalSndKill(int code, Datum arg)
634 Assert(MyWalSnd != NULL);
637 * Mark WalSnd struct no longer in use. Assume that no lock is required
641 DisownLatch(&MyWalSnd->latch);
643 /* WalSnd struct isn't mine anymore */
648 * Read 'nbytes' bytes from WAL into 'buf', starting at location 'recptr'
650 * XXX probably this should be improved to suck data directly from the
651 * WAL buffers when possible.
654 XLogRead(char *buf, XLogRecPtr recptr, Size nbytes)
656 XLogRecPtr startRecPtr = recptr;
657 char path[MAXPGPATH];
658 uint32 lastRemovedLog;
659 uint32 lastRemovedSeg;
669 startoff = recptr.xrecoff % XLogSegSize;
671 if (sendFile < 0 || !XLByteInSeg(recptr, sendId, sendSeg))
673 /* Switch to another logfile segment */
677 XLByteToSeg(recptr, sendId, sendSeg);
678 XLogFilePath(path, ThisTimeLineID, sendId, sendSeg);
680 sendFile = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
684 * If the file is not found, assume it's because the standby
685 * asked for a too old WAL segment that has already been
686 * removed or recycled.
690 char filename[MAXFNAMELEN];
692 XLogFileName(filename, ThisTimeLineID, sendId, sendSeg);
694 (errcode_for_file_access(),
695 errmsg("requested WAL segment %s has already been removed",
700 (errcode_for_file_access(),
701 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
702 path, sendId, sendSeg)));
707 /* Need to seek in the file? */
708 if (sendOff != startoff)
710 if (lseek(sendFile, (off_t) startoff, SEEK_SET) < 0)
712 (errcode_for_file_access(),
713 errmsg("could not seek in log file %u, segment %u to offset %u: %m",
714 sendId, sendSeg, startoff)));
718 /* How many bytes are within this segment? */
719 if (nbytes > (XLogSegSize - startoff))
720 segbytes = XLogSegSize - startoff;
724 readbytes = read(sendFile, buf, segbytes);
727 (errcode_for_file_access(),
728 errmsg("could not read from log file %u, segment %u, offset %u, "
730 sendId, sendSeg, sendOff, (unsigned long) segbytes)));
732 /* Update state for read */
733 XLByteAdvance(recptr, readbytes);
735 sendOff += readbytes;
741 * After reading into the buffer, check that what we read was valid. We do
742 * this after reading, because even though the segment was present when we
743 * opened it, it might get recycled or removed while we read it. The
744 * read() succeeds in that case, but the data we tried to read might
745 * already have been overwritten with new WAL records.
747 XLogGetLastRemoved(&lastRemovedLog, &lastRemovedSeg);
748 XLByteToSeg(startRecPtr, log, seg);
749 if (log < lastRemovedLog ||
750 (log == lastRemovedLog && seg <= lastRemovedSeg))
752 char filename[MAXFNAMELEN];
754 XLogFileName(filename, ThisTimeLineID, log, seg);
756 (errcode_for_file_access(),
757 errmsg("requested WAL segment %s has already been removed",
763 * Read up to MAX_SEND_SIZE bytes of WAL that's been flushed to disk,
764 * but not yet sent to the client, and send it.
766 * msgbuf is a work area in which the output message is constructed. It's
767 * passed in just so we can avoid re-palloc'ing the buffer on each cycle.
768 * It must be of size 1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE.
770 * If there is no unsent WAL remaining, *caughtup is set to true, otherwise
771 * *caughtup is set to false.
773 * Returns true if OK, false if trouble.
776 XLogSend(char *msgbuf, bool *caughtup)
778 XLogRecPtr SendRqstPtr;
782 WalDataMessageHeader msghdr;
785 * Attempt to send all data that's already been written out and fsync'd to
786 * disk. We cannot go further than what's been written out given the
787 * current implementation of XLogRead(). And in any case it's unsafe to
788 * send WAL that is not securely down to disk on the master: if the master
789 * subsequently crashes and restarts, slaves must not have applied any WAL
790 * that gets lost on the master.
792 SendRqstPtr = GetFlushRecPtr();
794 /* Quick exit if nothing to do */
795 if (XLByteLE(SendRqstPtr, sentPtr))
802 * Figure out how much to send in one message. If there's no more than
803 * MAX_SEND_SIZE bytes to send, send everything. Otherwise send
804 * MAX_SEND_SIZE bytes, but round back to logfile or page boundary.
806 * The rounding is not only for performance reasons. Walreceiver relies on
807 * the fact that we never split a WAL record across two messages. Since a
808 * long WAL record is split at page boundary into continuation records,
809 * page boundary is always a safe cut-off point. We also assume that
810 * SendRqstPtr never points to the middle of a WAL record.
813 if (startptr.xrecoff >= XLogFileSize)
816 * crossing a logid boundary, skip the non-existent last log segment
817 * in previous logical log file.
819 startptr.xlogid += 1;
820 startptr.xrecoff = 0;
824 XLByteAdvance(endptr, MAX_SEND_SIZE);
825 if (endptr.xlogid != startptr.xlogid)
827 /* Don't cross a logfile boundary within one message */
828 Assert(endptr.xlogid == startptr.xlogid + 1);
829 endptr.xlogid = startptr.xlogid;
830 endptr.xrecoff = XLogFileSize;
833 /* if we went beyond SendRqstPtr, back off */
834 if (XLByteLE(SendRqstPtr, endptr))
836 endptr = SendRqstPtr;
841 /* round down to page boundary. */
842 endptr.xrecoff -= (endptr.xrecoff % XLOG_BLCKSZ);
846 nbytes = endptr.xrecoff - startptr.xrecoff;
847 Assert(nbytes <= MAX_SEND_SIZE);
850 * OK to read and send the slice.
855 * Read the log directly into the output buffer to avoid extra memcpy
858 XLogRead(msgbuf + 1 + sizeof(WalDataMessageHeader), startptr, nbytes);
861 * We fill the message header last so that the send timestamp is taken as
864 msghdr.dataStart = startptr;
865 msghdr.walEnd = SendRqstPtr;
866 msghdr.sendTime = GetCurrentTimestamp();
868 memcpy(msgbuf + 1, &msghdr, sizeof(WalDataMessageHeader));
870 pq_putmessage('d', msgbuf, 1 + sizeof(WalDataMessageHeader) + nbytes);
872 /* Flush pending output to the client */
878 /* Update shared memory status */
880 /* use volatile pointer to prevent code rearrangement */
881 volatile WalSnd *walsnd = MyWalSnd;
883 SpinLockAcquire(&walsnd->mutex);
884 walsnd->sentPtr = sentPtr;
885 SpinLockRelease(&walsnd->mutex);
888 /* Report progress of XLOG streaming in PS display */
889 if (update_process_title)
891 char activitymsg[50];
893 snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
894 sentPtr.xlogid, sentPtr.xrecoff);
895 set_ps_display(activitymsg, false);
901 /* SIGHUP: set flag to re-read config file at next convenient time */
903 WalSndSigHupHandler(SIGNAL_ARGS)
907 SetLatch(&MyWalSnd->latch);
910 /* SIGTERM: set flag to shut down */
912 WalSndShutdownHandler(SIGNAL_ARGS)
914 walsender_shutdown_requested = true;
916 SetLatch(&MyWalSnd->latch);
920 * WalSndQuickDieHandler() occurs when signalled SIGQUIT by the postmaster.
922 * Some backend has bought the farm,
923 * so we need to stop what we're doing and exit.
926 WalSndQuickDieHandler(SIGNAL_ARGS)
928 PG_SETMASK(&BlockSig);
931 * We DO NOT want to run proc_exit() callbacks -- we're here because
932 * shared memory may be corrupted, so we don't want to try to clean up our
933 * transaction. Just nail the windows shut and get out of town. Now that
934 * there's an atexit callback to prevent third-party code from breaking
935 * things by calling exit() directly, we have to reset the callbacks
936 * explicitly to make this work as intended.
941 * Note we do exit(2) not exit(0). This is to force the postmaster into a
942 * system reset cycle if some idiot DBA sends a manual SIGQUIT to a random
943 * backend. This is necessary precisely because we don't clean up our
944 * shared memory state. (The "dead man switch" mechanism in pmsignal.c
945 * should ensure the postmaster sees this as a crash, too, but no harm in
946 * being doubly sure.)
951 /* SIGUSR1: set flag to send WAL records */
953 WalSndXLogSendHandler(SIGNAL_ARGS)
955 latch_sigusr1_handler();
958 /* SIGUSR2: set flag to do a last cycle and shut down afterwards */
960 WalSndLastCycleHandler(SIGNAL_ARGS)
962 walsender_ready_to_stop = true;
964 SetLatch(&MyWalSnd->latch);
967 /* Set up signal handlers */
971 /* Set up signal handlers */
972 pqsignal(SIGHUP, WalSndSigHupHandler); /* set flag to read config
974 pqsignal(SIGINT, SIG_IGN); /* not used */
975 pqsignal(SIGTERM, WalSndShutdownHandler); /* request shutdown */
976 pqsignal(SIGQUIT, WalSndQuickDieHandler); /* hard crash time */
977 pqsignal(SIGALRM, SIG_IGN);
978 pqsignal(SIGPIPE, SIG_IGN);
979 pqsignal(SIGUSR1, WalSndXLogSendHandler); /* request WAL sending */
980 pqsignal(SIGUSR2, WalSndLastCycleHandler); /* request a last cycle and
983 /* Reset some signals that are accepted by postmaster but not here */
984 pqsignal(SIGCHLD, SIG_DFL);
985 pqsignal(SIGTTIN, SIG_DFL);
986 pqsignal(SIGTTOU, SIG_DFL);
987 pqsignal(SIGCONT, SIG_DFL);
988 pqsignal(SIGWINCH, SIG_DFL);
991 /* Report shared-memory space needed by WalSndShmemInit */
993 WalSndShmemSize(void)
997 size = offsetof(WalSndCtlData, walsnds);
998 size = add_size(size, mul_size(max_wal_senders, sizeof(WalSnd)));
1003 /* Allocate and initialize walsender-related shared memory */
1005 WalSndShmemInit(void)
1010 WalSndCtl = (WalSndCtlData *)
1011 ShmemInitStruct("Wal Sender Ctl", WalSndShmemSize(), &found);
1015 /* First time through, so initialize */
1016 MemSet(WalSndCtl, 0, WalSndShmemSize());
1018 for (i = 0; i < max_wal_senders; i++)
1020 WalSnd *walsnd = &WalSndCtl->walsnds[i];
1022 SpinLockInit(&walsnd->mutex);
1023 InitSharedLatch(&walsnd->latch);
1028 /* Wake up all walsenders */
1034 for (i = 0; i < max_wal_senders; i++)
1035 SetLatch(&WalSndCtl->walsnds[i].latch);
1038 /* Set state for current walsender (only called in walsender) */
1040 WalSndSetState(WalSndState state)
1042 /* use volatile pointer to prevent code rearrangement */
1043 volatile WalSnd *walsnd = MyWalSnd;
1045 Assert(am_walsender);
1047 if (walsnd->state == state)
1050 SpinLockAcquire(&walsnd->mutex);
1051 walsnd->state = state;
1052 SpinLockRelease(&walsnd->mutex);
1056 * Return a string constant representing the state. This is used
1057 * in system views, and should *not* be translated.
1060 WalSndGetStateString(WalSndState state)
1064 case WALSNDSTATE_STARTUP:
1066 case WALSNDSTATE_BACKUP:
1068 case WALSNDSTATE_CATCHUP:
1070 case WALSNDSTATE_STREAMING:
1078 * Returns activity of walsenders, including pids and xlog locations sent to
1082 pg_stat_get_wal_senders(PG_FUNCTION_ARGS)
1084 #define PG_STAT_GET_WAL_SENDERS_COLS 3
1085 ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1087 Tuplestorestate *tupstore;
1088 MemoryContext per_query_ctx;
1089 MemoryContext oldcontext;
1092 /* check to see if caller supports us returning a tuplestore */
1093 if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
1095 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1096 errmsg("set-valued function called in context that cannot accept a set")));
1097 if (!(rsinfo->allowedModes & SFRM_Materialize))
1099 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1100 errmsg("materialize mode required, but it is not " \
1101 "allowed in this context")));
1103 /* Build a tuple descriptor for our result type */
1104 if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1105 elog(ERROR, "return type must be a row type");
1107 per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
1108 oldcontext = MemoryContextSwitchTo(per_query_ctx);
1110 tupstore = tuplestore_begin_heap(true, false, work_mem);
1111 rsinfo->returnMode = SFRM_Materialize;
1112 rsinfo->setResult = tupstore;
1113 rsinfo->setDesc = tupdesc;
1115 MemoryContextSwitchTo(oldcontext);
1117 for (i = 0; i < max_wal_senders; i++)
1119 /* use volatile pointer to prevent code rearrangement */
1120 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1121 char sent_location[MAXFNAMELEN];
1124 Datum values[PG_STAT_GET_WAL_SENDERS_COLS];
1125 bool nulls[PG_STAT_GET_WAL_SENDERS_COLS];
1127 if (walsnd->pid == 0)
1130 SpinLockAcquire(&walsnd->mutex);
1131 sentPtr = walsnd->sentPtr;
1132 state = walsnd->state;
1133 SpinLockRelease(&walsnd->mutex);
1135 snprintf(sent_location, sizeof(sent_location), "%X/%X",
1136 sentPtr.xlogid, sentPtr.xrecoff);
1138 memset(nulls, 0, sizeof(nulls));
1139 values[0] = Int32GetDatum(walsnd->pid);
1143 * Only superusers can see details. Other users only get
1144 * the pid value to know it's a walsender, but no details.
1151 values[1] = CStringGetTextDatum(WalSndGetStateString(state));
1152 values[2] = CStringGetTextDatum(sent_location);
1155 tuplestore_putvalues(tupstore, tupdesc, values, nulls);
1158 /* clean up and return the tuplestore */
1159 tuplestore_donestoring(tupstore);
1165 * This isn't currently used for anything. Monitoring tools might be
1166 * interested in the future, and we'll need something like this in the
1167 * future for synchronous replication.
1171 * Returns the oldest Send position among walsenders. Or InvalidXLogRecPtr
1175 GetOldestWALSendPointer(void)
1177 XLogRecPtr oldest = {0, 0};
1181 for (i = 0; i < max_wal_senders; i++)
1183 /* use volatile pointer to prevent code rearrangement */
1184 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1187 if (walsnd->pid == 0)
1190 SpinLockAcquire(&walsnd->mutex);
1191 recptr = walsnd->sentPtr;
1192 SpinLockRelease(&walsnd->mutex);
1194 if (recptr.xlogid == 0 && recptr.xrecoff == 0)
1197 if (!found || XLByteLT(recptr, oldest))